Intro

Today, OpenAI's new model, Astra, solved ten unsolved mathematical problems using only $2,000 worth of compute, practically backing scientists into a corner! I'm Muyan, and welcome to the AI Daily, produced by Mark's Tech Insights team. In addition to this battle of wits, today we'll also talk about the inside story of Apple quietly joining forces with Alibaba, and the EU's heavy-handed new regulations officially targeting AI today. Stay tuned!

Today's Top Stories

1. OpenAI Releases New Model Astra, Solving Ten Unsolved Math Problems with Just $2,000 in Compute

• Source: AI Weekly / imfounder.com (https://aiweekly.co/ai-news-today）
• Summary: An internal version of OpenAI's brand-new reasoning model, Astra, has successfully solved ten open problems in theoretical computer science and mathematics, publishing verifiable Lean formal proofs on GitHub. Most shockingly, the entire computation process cost only about $2,000 in compute.
• Most Surprising Aspect: Problems that previously required top mathematicians years or even a lifetime to study can now be solved instantly by AI for just $2,000 in electricity and compute, even directly writing flawless mathematical proofs.
• Taiwan Perspective: This indicates that "reasoning capability" has been completely commoditized and made affordable. Taiwan's semiconductor and server supply chains should pay close attention: future market demand for high-performance, low-cost reasoning chips will experience explosive growth.
• Discussion Points:
  1. With reasoning costs dropping so low, will this accelerate the "de-humanization" of scientific research?
  2. What will the future role of mathematicians be? Will they only be responsible for "posing questions" and "verification"?
• Script Suggestion: You might think math puzzles are far removed from our daily lives, but think about it: problems that top professors racked their brains over but couldn't solve are now thrown at Astra by OpenAI. For just $2,000USD\text{—}about60,000NTDinelectricity\text{—}itcrackedthemall,andevenputtheproofsonGitHub!Whatdoesthismean?The\mathrm{"}reasoningability\mathrm{"}ofhigherintelligencehasbecomeacommodityascheapastapwater\mathrm{.}Inthefuture,whenstudentsinTaiwanstudymath,should{n}^{\mathrm{\prime }}ttheythinkmoreabouthowtocollaboratewiththis\mathrm{"}2,000\; USD—about\; 60,000\; NTD\; in\; electricity—it\; cracked\; them\; all,\; and\; even\; put\; the\; proofs\; on\; GitHub!\; What\; does\; this\; mean?\; The\; "reasoning\; ability"\; of\; higher\; intelligence\; has\; become\; a\; commodity\; as\; cheap\; as\; tap\; water.\; In\; the\; future,\; when\; students\; in\; Taiwan\; study\; math,\; shouldn\text{'}t\; they\; think\; more\; about\; how\; to\; collaborate\; with\; this\; "$2,000 superbrain" instead of going head-to-head with it?

---

2. Breakthrough of the Century! Anthropic Mathematician Uses Claude Fable 5 to Disprove the Century-Old "Jacobian Conjecture"

• Source: imfounder / AI Updates August 2026 (https://imfounder.com/science-tech/ai/ai-updates-august-2026-openai-astra-deepmind/）
• Summary: A mathematician working at Anthropic used the company's latest Claude Fable 5 model to successfully find a counterexample, completely disproving the "Jacobian Conjecture," which had remained unsolved in algebraic geometry for nearly 90 years. If ultimately verified by the academic community, this discovery will become one of the most influential mathematical achievements of this century.
• Most Surprising Aspect: AI is not just helping with tedious calculations; it is exhibiting genuine "creative intuition" by proactively finding a fatal counterexample that generations of human mathematicians failed to locate.
• Taiwan Perspective: Taiwan's tech and education sectors should rethink their approach. When even pure theoretical mathematics like algebraic geometry can be overturned by AI, should our higher education academic research methods fully integrate AI collaboration? Otherwise, we simply won't be able to compete internationally in terms of research efficiency.
• Discussion Points:
  1. Disproving a conjecture requires "counter-intuitive" creativity. Does this prove that AI already possesses some form of creative thinking?
  2. In the future, is it possible for the Nobel Prize or Fields Medal to be awarded to an AI collaborator?
• Script Suggestion: Right after talking about OpenAI, their arch-rival Anthropic delivered a slap in the face! One of their mathematicians used Claude Fable 5 to actually disprove the "Jacobian Conjecture," which has baffled the math world for nearly 90 years. This is truly mind-blowing because while "proving" something exists is hard, imagining a "counterexample" out of thin air to overturn a theorem everyone believed in requires immense creativity. It's like being in a maze where humans searched for 90 years thinking there was no exit, and the AI just walked straight through the wall to build a new path. Now, who dares to say AI can only mimic and lacks creativity?

---

3. OpenAI Slashes GPT-5.6 Luna Price by 80%, Opening It to All Free Users with No Character Limits

• Source: Tech Insider / AIToolsReview (https://tech-insider.org/ca/gpt-5-6-luna-free-price-cut-2026/）
• Summary: OpenAI has slashed the API pricing for its flagship model, GPT-5.6 Luna, by a massive 80%, bringing the cost per million tokens down to an ultra-low $0.2to0.2\; to$1.2. At the same time, OpenAI has made this model the default for all free ChatGPT users and completely removed character limits on conversations.
• Most Surprising Aspect: A top-tier model that required a paid subscription just a few weeks ago is now being handed out for free with no character limits. This AI price war has turned into an absolute bloodbath.
• Taiwan Perspective: This is a massive boon for Taiwan's small and medium software startups! Complex applications that they didn't dare develop before due to high API costs can now be built on a shoestring budget, presenting a golden opportunity to expand into international markets.
• Discussion Points:
  1. Is OpenAI bleeding cash like this to completely monopolize the market, or do they have much cheaper compute backing them up?
  2. Now that the free version has no character limits, is there any survival space left for third-party wrapper tools?
• Script Suggestion: Ladies and gentlemen, OpenAI just flipped the table today! They slashed the API price of their latest and greatest GPT-5.6 Luna by a whopping 80% just like that! What's even crazier is that you no longer need to pay the $20 monthly subscription fee—the free version of ChatGPT now defaults to this powerful brain, and there's "no character limit"! This is practically selling gold at the price of cabbage. I think OpenAI is using a cutthroat price war to corner other small and medium AI startups. This "kill a thousand enemies at the cost of eight hundred of your own" strategy is incredibly ruthless.

---

4. Apple Secretly Partners with Alibaba to Train China-Exclusive AI Model, Becoming First Foreign Firm to Win Beijing's AI Approval

• Source: MacRumors / Japan Times (https://www.macrumors.com/2026/08/14/apple-trained-own-ai-model-for-china/）
• Summary: Apple has partnered with Alibaba to train a large language model specifically for the Chinese market, which has successfully passed Chinese government regulatory reviews. After previous negotiations with giants like Baidu, ByteDance, and Tencent fell through, this collaboration with Alibaba finally paves the way for Apple Intelligence to reach hundreds of millions of iPhone users in China.
• Most Surprising Aspect: To enter the Chinese market, Apple—normally so privacy-focused and fiercely independent—actually chose to team up with a local Chinese e-commerce giant and even passed strict official regulatory reviews.
• Taiwan Perspective: This is undoubtedly a shot in the arm for Taiwan's Apple supply chain. If iPhones in the Chinese market can successfully ship with AI features, it will help boost sales, thereby driving momentum for Taiwanese manufacturers like TSMC and Largan Precision.
• Discussion Points:
  1. How many compromises on privacy and response content will be made for an AI specifically "custom-tailored" for China and approved by its government?
  2. Will this become a blueprint for other foreign tech companies looking to enter China?
• Script Suggestion: There really are no permanent enemies in the business world. To save its sales in the Chinese market, Apple is really going all out this time! After being turned down by Baidu and Tencent, they quietly partnered with Alibaba to train a "China-exclusive" AI, and even secured an official license from Beijing, making them the first foreign tech company to clear this hurdle. However, I'm also very curious: how will this strictly vetted Apple Intelligence respond when asked about sensitive topics? Will it suddenly become very "politically savvy"?

---

5. Boston Dynamics' Atlas Humanoid Robot Enters Hyundai Factories, Powered by Google's Gemini Robotics Model

• Source: Boston Dynamics / The Robot Report (https://bostondynamics.com/blog/boston-dynamics-google-deepmind-form-new-ai-partnership/）
• Summary: The all-electric version of the Atlas humanoid robot has officially moved past the lab and stunt-show phase, entering Hyundai Motor's production lines for actual industrial deployment. Powered by Google DeepMind's Gemini Robotics foundation model, all production capacity for 2026 has already been fully booked by Hyundai and Google.
• Most Surprising Aspect: Those viral robots that used to do backflips and dance on YouTube are now actually putting on work uniforms and heading to factories for a day job.
• Taiwan Perspective: As a global hub for automation equipment and precision components, Taiwan should keep a close eye on this wave of "Embodied AI" deployment. From motors and sensors to reducers, Taiwanese manufacturers definitely have massive opportunities to break into the humanoid robot supply chain.
• Discussion Points:
  1. When robots are no longer just performing mechanical repetitions but possess a Gemini brain, how much of an impact will this have on the blue-collar job market?
  2. Will this wave of humanoid robot commercialization trigger a new hardware arms race?
• Script Suggestion: We used to watch Boston Dynamics' Atlas robot do backflips and think of it as a high-tech toy. But now, it's actually "punching the clock" at a factory! On Hyundai's production lines, these all-electric Atlas robots, powered by Google's Gemini brain, are starting to do real physical labor. This means the era of "Embodied AI" has officially arrived. I'm thinking, in the future when we tour factories, we might not see rows of cold robotic arms, but humanoid robots greeting each other and assembling parts while thinking with AI. Just imagining that scene feels incredibly sci-fi!

---

6. EU AI Act Enforcement Begins: All Chatbots in Europe Must Proactively Declare "I am an AI"

• Source: Cooley / European Commission (https://www.cooley.com/news/insight/2026/2026-08-03-eu-ai-act-transparency-obligations-take-effect-2-august-2026）
• Summary: Article 50 of the EU AI Act officially took effect on August 2. The new regulation mandates that all chatbots must explicitly disclose their AI identity, deepfake content must be labeled, and AI-generated content must carry machine-readable watermarks. Violators face fines of up to €15 million or 3% of their global annual turnover.
• Most Surprising Aspect: The EU is playing for keeps this time. Not only are the fines staggering, but they have also directly established an "AI Office" for cross-border oversight. The "Wild West" era of AI regulation is officially over.
• Taiwan Perspective: Many Taiwanese software teams developing apps or SaaS for the European market must immediately review their product compliance, especially whether their user interfaces (UI) include clear AI disclosures. Otherwise, they could easily cross the red line and face astronomical fines.
• Discussion Points:
  1. Will mandatory AI labeling reduce user trust, or will it actually increase transparency?
  2. Will such strict regulations stifle the innovation of local European AI startups, causing talent to drain to the Americas and Asia?
• Script Suggestion: The EU's heavy fist has finally come down! Now in Europe, if your app has a chatbot, it has to be completely honest and declare "I am an AI" right off the bat. If you use AI to make deepfake videos or write articles without adding machine-readable watermarks, sorry, but a fine of up to €15 million or 3% of your global turnover will be sent straight to your door. This is a massive wake-up call for Taiwanese startup teams looking to sell products in Europe. Compliance costs are now more important than ever—don't play games with your wallet!

Closing Thoughts

Today, from OpenAI's affordable reasoning brain to Apple's survival story in China, and finally to the EU's regulatory hammer, we are standing right in the eye of the storm where AI technology deeply collides with the real world. I'm Muyan. Thank you for listening to today's Mark's Tech Insights. See you next time, bye-bye!